AI-Powered Drones Have Killed People in Ukraine, Risks of Cyberattacks Loom

AI-powered drones have already killed people in Ukraine, highlighting the immediate dangers of AI in conflict scenarios. Grace Huckins notes that AI-driven cyberattacks on hospitals are expected to claim victims in the future, raising alarms about the potential for AI agents to execute attacks on critical infrastructure.

Speculative risks associated with AI include the possibility of AI-designed pathogens that could significantly impact the population and scenarios where AI could lead to a world economic crash, resulting in conflicts and famine. Will Douglas Heaven asserts that there are no circumstances outside of apocalyptic science fiction in which AI could kill us all, emphasizing the current limitations of AI alignment research.

Current AI models are inconsistent and unpredictable, and AI companies have not yet developed fully aligned models, which complicates monitoring efforts. The existing approaches to monitor AI agents are described as fragile. Furthermore, the US government has not intervened to regulate AI companies, leaving a regulatory gap.

This follows a July open letter signed by employees urging AI firms to focus on alignment and a slowdown in development. The conversation around AI risks is further complicated by the conflict of interest inherent in self-regulation by AI companies. The recent hack incident involving Hugging Face, analyzed by METR, underscores the vulnerabilities within AI platforms.

Overall, the discourse around AI risks is evolving, with a clear need for more robust regulatory frameworks and alignment strategies to mitigate potential dangers.